Intel copies Arduino

The PS (Processing System) and PL (Programmable Logic) are connected with a 32 bits AXI switch.

Zynq's "unlimited bandwidth" is subject to the AXI interconnect General Purpose DMA interface (AXI_GP) maximum rate, which is 600MBytes/sec or 4.8Gbps.

Quark has two 2.5Gbps PCIe, or total of 5.0Gbps of "unlimited bandwidth".

So, they are pretty much neck-to-neck. I guess the chip designers are reading each other's spec.

You mean like some old 80286 PC AT? Eat floppy disks? Probably thousands would fit in a modern ASIC.

Back to the 80's.

Absolutely. It is quite telling that all the big players have been forced to introduce ARM products that must surely compete directly with their own product lines.

I guess there is still a lot of inertia in some markets, automotive engine controllers say.

--

John Devereux

Den søndag den 3. november 2013 16.52.09 UTC+1 skrev snipped-for-privacy@gmail.com :

o you later.

a 32 bits AXI switch.

urpose DMA interface (AXI_GP) maximum rate, which is 600MBytes/sec or 4.8Gb ps.

per channel per direction and that is only one interface, it has more

afaict the zynq supports gen2 PCIe, 5Gbps per port

ading each other's spec.

-Lasse

On Sunday, November 3, 2013 11:16:08 AM UTC-8, Lasse Langwadt Christensen w rote:

om:

to you later.

cy

th a 32 bits AXI switch.

Purpose DMA interface (AXI_GP) maximum rate, which is 600MBytes/sec or 4.8 Gbps.

In this context, we are only comparing Processor to FPGA wrt:

1. Zynq: 32 bits AXI switch (to internal FPGA)
2. Quark: x2 PCIe (to external PCIe FPGA)

Yes, Zynq can transfer to memory at higher speed and wider path (64 bits), but not to FPGA.

".

Yes to external PCIe device, but not to internal FPGA.

Inertia, sure, but the writing is on the wall. Where every penny counts, the obsolete can't be tolerated long.

It already is, IME, at least for things that need more than 8051/PIC/AVR levels of horsepower.

As a software guy, the thing I don't like about ARM for embedded is that the tool vendors seem to assume that if you've chosen an ARM CPU, you must be doing consumer devices in huge volumes, and can therefore amortize a $10,000 compiler or $2,500 debugger over hundreds of thousands of units. I don't usually have to *pay* for the tools, but if the tool the *boss* paid for turns out to suck, he or she would rather have me code around it than try to get approval for another big purchase.

At a previous job, the hardware guys had a whole stack of a certain model of ARM debugger. They would burn out on a regular basis, and when there were enough dead ones in the pile, the boss would send the whole pile back at once to get fixed. The thing initially sold for a few thousand dollars, and there was a required support contract on it to the tune of several hundred per year; if you let the support contract lapse and then tried to turn it back on later, they'd charge you a penalty for the time it was lapsed. This was all considered normal by the hardware guys.

If the design can handle the power requirements, Atom is (or was?) sort of interesting for a smaller shop that wants to use Linux as the OS, because you can get 95-99% of the chip lit up quick with a desktop Linux distribution. Adding binary-only drivers from Intel often gets you the rest of the way. If you can't afford the light bill, or if you are big enough to have (or to rent) people that can port the OS for you, then you probably pick ARM.

Another split is a straight ARM CPU, vs something with ARM and a DSP in one can (TI). TI has cheap or free development tools and libraries, but you need approximately 37 of them to light up one of their chips 100%, and the right order to connect those 37 things has to be found by experiment. There is also only one correct way to get data into and out of the DSP side, and hundreds of wrong ways. If you don't have time to figure all that out, TI will be happy to refer you to one of their development partners, who have connected all 37 things and know the one way and have a reference design. If what you want differs a lot from the reference design, be prepared to fork over lots of cash, and not get what you want. (For extra fun, many of the partners are not in the US, and it only takes a few months to get a copy of the reference hardware.)

...and that have sane development environments. In the short term, if the tools only work in 2009-era Ubuntu, then it's not hard to keep a VM of that around - for now. In 10 years, if a 32-bit VM gets hard to come by, it might be more interesting.

Matt Roberds

You don't pick one set of tools and stick with it? That's the mode I see for ARM development. Well, other than us hardware guys who get by with the freebies. I hear the opposite from the software types. The tools are chosen by the leads and they have to make them work. Management buys what the experienced software guys want. Once that decision has been made, it's tough to unmake it, though.

Huh? I've never heard of such a thing. Well, at least not since the early days of x86 and 68K, when the tools were tens of thou$ands and HP was in the business. All of our hardware tools are either given to is by the supplier or cost If the design can handle the power requirements, Atom is (or was?) sort

Ick. That Intel stuff is EXPENSIVE. OS? We're talking M0s here. Sure, an A9 would probably be using an OS but Linux is readily available for it and there are others (expensive for a small shop, though).

You'll pay an arm and a leg for the power solution, forgetting the CPU itself. We use TIs all the time. The power requirements are a mess. I wear, they're in the CPU business to sell PMICs.

Or 30 minutes... Cash for a reference DESIGN? Huh?

ehh? what's wrong with GCC?

-Lasse

Another big advantage of the Intel board is that the A/D converters are

12 bit, rather than 10 bit A/D converters on the Arduino boards. I wanted to use an Arduino board for something where a 10 bit A/D was not enough, we needed an 11 or 12 bit.

Only Windows people pay that kind of money for tools.

As a Windows user/developer I like free as well.

Atmel M3/M4 processors have a free compiler:

The USB debugger SAM-ICE is about $100:

Atmel has many eval kits too:

So, maybe a few hundred dollars, but you get started and working in little time.

Good Luck

They can sell the business to Marvell again. I remember working with Intel's embedded stuff back in the 1980's with those big blue development systems with 8 inch floppy drives.

Intel realized that the most profitable use of a fab was to manufacture high-margin x86 chips and it's tough to get away from that philosophy. Now they have no choice.

is interesting.

On a sunny day (Mon, 04 Nov 2013 11:58:28 -0800) it happened sms wrote in :

IIRC I did read rcently that Intel wll be making ARM chips in its fabs.

Forget it - Atmel Studio runs only under Microsoft Visual Studio, and the whole kit and kaboodle is a monster. Also, Atmel's hardware setup code is as heavy as Windows binaries.

(I made the mistake of installing Atmel Studio, as their development boards have a non-standard bastardized Segger J-Link JTAG interface, which does not work without Atmel Studio.)

My recommendation is something GCC -based, e.g. Yagarto.

For an IDE, Eclipse is the way to go.

For debugging, OpenOCD and GDB will do the job. There are many graphical front-ends for GDB, if the plain user interface one is not attractive enough.

--

Tauno Voipio

ST cortex M4:

includes on board debugger via USB

setting up eclipse:

$14.90 and you are ready to go

$7.99 if you only need the M0

-Lasse

-Lasse

According to more than one manager I have worked for, the problem with gcc is that there is nobody to sue if it goes wrong. When I point out that suing people does not write code or cause products to ship on time, my protests usually fall on deaf ears. I think "availability of someone to sue" is an insane criterion for tool selection, but I'm usually not the person making the decision.

Matt Roberds

PS Please consider using something other than Google Groups. It makes an incredible mess out of followups, which then has to be manually corrected.

The company picks out one set and sticks with it. Forever. 5 or 10 years after purchase, the tools are starting to not support more recent CPUs. When "not support" means "doesn't work at all", they will buy upgrades or switch tools, but when it means "generates code that works, but is slow, doesn't take advantage of the new features of the CPU, and is harder to write and debug", they won't buy upgrades. They think programmer time is cheaper than tools.

That's what I am talking about. Sometimes when even the experienced guys complain that the current tool isn't getting it done anymore, they can't get manglement approval to change.

I'm going by what the hardware guys told me, and what I found when I Googled that debugger (because I didn't believe the cost and support contract thing either). The pricing and terms on the web site matched what they said. I'm trying and failing to remember who made it; I remember it wasn't ARM itself. This was for Cortex-Ax series CPUs.

The particular product I knew about that used Atom had enough margin to cover it. (It was price-competitive with the competition - maybe the competition was using ARM and making more money, though...)

Ah. I am talking about Cortex-A type stuff (ARMv7). Something you'd use as the main CPU in a tablet or smartphone.

Patented software guy response #3: Sounds like a hardware problem to me! :) As long as it boots up and runs, I don't really care how many of those little square black things you guys have to use. (Of course, too many little square black things means it's expensive, runs hot, sucks battery, doesn't fit in the hole, doesn't sell, everyone gets laid off. How embarrassing.)

I was sort of summarizing several things quickly. Here it is in more detail. Also note that this was dealing with the higher-end TI video processors with ARM and DSP in one can.

The reference *design* (on paper) is free, either from TI or from one of their partners.

The *demo board* (hardware) that implements the reference design is very not free, and ships from one of TI's partners, usually in Asia, usually very slowly. (This wasn't an issue of not wanting to pay for overnight shipping; the story was that there simply weren't any to be had yet.)

If what you want is close to the reference design, and you just want blue LEDs instead of green, then OK. Change the BOM and call your favorite PCB factory and get going.

If what you want is far different from the reference design, and you are a small company without enough resources to redesign it yourself, TI won't talk to you directly; you pretty much *have* to go through one of their partners to get design help.

At least one of said TI partners will take your spec, throw away all the parts of it they don't understand, badly implement the parts they do understand, and charge you a lot of money for it all.

The only thing worse than doing it yourself is paying somebody else to do it for you, and vice versa. :)

Matt Roberds

You mean there are compiler vendors that do not "disclaim all liability"?

--

John Devereux

That manager was pulling your leg, he did not have an answer for you so he made one up.

Some managers think the more money they spend in developing a product, means they are doing a better job.

Upper management does not know either, so spending money is their only basis to believe that the rank-n-file is doing a good job.

As the rank-n-file, we know it does not matter if the compiler is free or high priced.

So if your manager bows to the "higher price is better" mentality, show them this article from another thread:

This article never mentions compilers.

But it helps make clear, time and consistent coding practice will rule the day.

hamilton